Agents Accelerate Faster Than Their Guardrails
from AI & Tech Daily with Jesse Owen Hosted by Jesse Owen
Anthropic says Claude now leads 26% of its model research and development, putting a sharper number on AI helping to build its successors. We examine the limits of that claim, the monitoring burden created by 30,000 concurrent internal agents, a new embedded-evaluation deal with Accenture and California's push for independently tested safeguards. Also: sovereign Gemma 4 inference on AWS, MLPerf's first end-to-end RAG and edge-agent tests, active exploitation of exposed Vite development servers, Google's experimental household agent, and the open-source Mantis security harness. Full transcript: https://www.owenonthenet.com/podcast/2026-09-19/ Sources: https://www.anthropic.com/institute/measuring-pace-of-ai-development https://apnews.com/article/anthropic-claude-ai-model-self-improvement-4d3a7430f57cbc7c39e1c5f2b7d7e132 https://www.anthropic.com/news/accenture-embedded-evaluation https://www.gov.ca.gov/2026/09/18/governor-newsom-issues-executive-order-to-accelerate-independent-oversight-and-advance-the-creation-of-an-ai-kill-switch https://aws.amazon.com/blogs/security/run-open-weight-models-on-aws-bedrock-in-aws-european-sovereign-cloud https://mlcommons.org/2026/09/mlperf-inference-v6-1-results https://www.csa.gov.sg/alerts-and-advisories/alerts/al-2026-124 https://github.com/vitejs/vite/security/advisories/GHSA-v2wj-q39q-566r https://blog.google/innovation-and-ai/models-and-research/google-labs/cc-expanding-to-groups https://cloud.google.com/blog/topics/systems/using-ai-agents-to-secure-google-infrastructure https://github.com/google/mantis
Embed this episode
What this episode covers
Anthropic says Claude now leads 26% of its model research and development, putting a sharper number on AI helping to build its successors. We examine the limits of that claim, the monitoring burden created by 30,000 concurrent internal agents, a new embedded-evaluation deal with Accenture and California's push for independently tested safeguards. Also: sovereign Gemma 4 inference on AWS, MLPerf's first end-to-end RAG and edge-agent tests, active exploitation of exposed Vite development servers, Google's experimental household agent, and the open-source Mantis security harness.
No transcript yet.